Diagnosis Of Coronary Artery Calcification In Type 2 Diabetes Patie Nts Using U-NET Model Based On Lung CT Scanning

Purpose:To verify the reliability and clinical feasibility of artificial intelligence model in predicting the degree of coronary calcification on conventional chest computed tomography in patients with type 2 diabetes.Methods:By using the electronic medical record retrieval system of special disease bank and general disease bank,routine chest CT and coronary artery CTA were selected from January 2010 to June 2022 in Dalian Central Hospital A total of 225 patients with type2 diabetes who were older than 18 years and had clear coronary artery calcification score in coronary artery CTA were examined Conventional chest computed tomography images were used as input values and U-NET was used as the backbone network.Dense Net was the deep learning network structure of the encoder as the model to perform automatic calcium quantization,and the output value was the coronary artery calcium score(CACS),which divided the calcifications into class 0,Class 1,and class 2calcifications(class 0 was<100,Class 1 was 100-calcium score f400,and class 2 was calcium score>400).Accuracy,Precision,Recall,Specificity,Negative predictive value and F1 were used to evaluate the diagnostic efficiency of U-NET model for each CACS category.The Agatston score calculated manually by radiologists with more than 5years of experience was used as the reference standard.SPSS25.0 software was used to analyze the normality of Al-CACS and manually measured gold standard CACS.The paired sample rank sum test was used to compare the coronary artery calcification scores measured by the two methods in the same patient.Intra-class correlation coefficient(ICC)analysis,Spearman analysis and Bland-Altman analysis were used to compare the consistency between the AI predicted CAC score and the manually measured gold standard CAC score.And the CAC risk category assessment was carried out.Weighted Kappa analysis was used to quantify the correlation between the CAC risk category predicted by artificial intelligence and the gold standard category measured manually.Results:This study included 225 patients with type 2 diabetes as the data set.1.Diagnostic efficiency of U-NET model for each coronary calcification score category.the model reaches accuracy 0.91,precision 0.95,recall 0.89,specificity0.93,negative predictive value 0.85,F1 0.92 in category O-light,accuracy 0.87,precision0.61,recall 0.84,specificity 0.87,negative predictive value 0.96,F1 0.71 in 1-moderate and accuracy 0.96,precision 0.97,recall 0.82,specificity 0.99,negative predictive value0.96,F1 0.89.2.Comparison of coronary artery calcium score predicted by U-NET model and gold standard coronary artery calcium score.Paired sample rank sum test showed no difference between the artificial intelligence model and the gold standard CAC score predicted by manual measurement(P=0.069>0.05).3.Consistency test between U-NET model predicted coronary artery calcium score and gold standard coronary artery calcium score.The artificial intelligence prediction model achieved a good agreement with the gold standard on the CAC score prediction,and the ICC reached 0.89(95%Cl:0.86.0.92.p<0.01).Spearman’s correlation coefficient ρ was achieved 0.93(95%Cl:0.910.94p<0.01);In the Bland-Altman plotit can be clearly seen that the mean difference between the entry intelligence model prediction and the gold standard integral measured manually is 41.5(95%C1:-362.1.445.2)4.Assessment of risk categories for coronary artery calcificatior predicted by the U-NET model.The current model has also achieved good results in the classification of CAC with the three-classification Accuracy reaching 87%and Kappa achieving 0.77(95%C:0.69.0.84),AI prediction model algorithm,a total of 195 people were correctly classified by AI prediction model 128 people,59 people and 38 people were classified into class 0 calcification,Class 1calcification and class 2 calcification respectively(class0 was<=100.1 class 100<calcification score<=400.2 Class calcification score>400).Conclusion:What model of artificial intelligence can accurately identify and quantify coronary artery calcification and its risk stratification in lung CT scanning technology for patients with type 2 diabetes mellitus,which can obviously improve the work efficiency of medical workers and has certain clinical value,making it possible for artificial intelligence to enter the clinic.